School of Engineering Department of Computer Science and Engineering 96 Federated Learning in Healthcare with Privacy-Preserving Supervisor: CHEN Hao / CSE Student: WONG Hoi Tin / FINA Course: UROP 2100, Fall This project aims to conduct comprehensive research on federated learning (FL) frameworks and related works in healthcare, especially in medical image recognition. The existing problem of domain shift is critical in medical image recognition in federated settings (Ying et al., 2023). Thus, this my research work will be focusing on domain generalization (DG) in FL. This progress report covers the first part of the project, problem formulation and research on related work has been finished in this stage. Meanwhile, the prototyping of p2p architecture, personalization and generalization methods of my proposed framework, p2p-FedIOG, have been completed, trials on PACS datasets have proven the feasibility of the prototype. The last part of the report will present preliminary experiment settings for next stage. Multimodal Learning for Cancer Diagnosis and Prognosis Supervisor: CHEN Hao / CSE Student: CHOW Chung Yan / CPEG Course: UROP 1100, Fall UROP 2100, Spring UROP 3200, Summer This project is the continuation of the model outlined in the UROP 2100 report. This report will focus on the experiment results as well as the changes in the model architecture. Suggestions on further improvement are partially implemented. Multimodal Learning for Cancer Diagnosis and Prognosis Supervisor: CHEN Hao / CSE Student: HOU Jingcheng / COMP Course: UROP 1100, Summer Developing large-scale vision-language models (LVLM) has shown enormous potential across various tasks. However, the shortage of comprehensive image-text data medicine has slowed the pace of LVLM development in this area. In this report, several advanced medicine LVLMs will be chosen to be analyzed and compared in terms of architecture and data preparation methods. Besides, we will also present our progress in the evaluation dataset annotation and cleaning in MedDr. Future work will also be discussed.
RkJQdWJsaXNoZXIy NDk5Njg=